Present day electronic and semiconductor integrated circuits are typically built on a multi-layer substrate or board which is used to interconnect the electronic components or devices which comprise the circuit. The multi-layer boards or substrates, in turn, consist of planar conductive layers separated by planar insulating layers. Portions of some of the conductive layers may be removed leaving electrically conductive signal lines or "traces" which connect the components on a particular layer. Conductive traces in different layers are typically electrically connected by forming holes in the intervening layers and plating the inside surfaces of the holes to form structures called "plated-through vias." Direct current power and ground are also provided to the components and devices in one or more of the layers.
Signals in digital electronic systems typically carry information by alternating between two voltage levels (i.e., a low voltage level and a high voltage level) so that a digital signal, over time, has an alternating voltage characteristic with an average frequency. Digital electronic systems are continually being produced which operate at higher and higher signal frequencies. The requirements of a power-distribution network which operates with high frequency circuits include a low impedance in the conductors from the power supply to the circuit over the frequency range at which the circuit operates and a low-pass filter-type transfer response which does not peak in the frequency range in which the circuit operates.
The low supply impedance is necessary because components and devices operating at high frequencies induce high frequency transient load currents in the conductors carrying power to the components. At the high frequencies present in the power conductors, any appreciable impedance can cause a significant drop in the voltage present at the component. The supply impedance usually has a design or target value which is calculated from the amount of noise generated by the electronics.
The requirement for low-pass type transfer impedance without peaking arises from the fact that DC power must be transmitted on the power-distribution network, but noise should not be transmitted. The bandwidth of such noise is usually equal to, or higher than, the bandwidth of signals that must be transmitted on the signal traces, so that the noise can be reduced by a low-pass type transfer impedance. In prior art analog circuits and digital electronics, the power-distribution network consisted of wires, metal bus bars, traces and bypass capacitors placed close to the electronic components and devices. This kind of structure inherently provides a low-pass characteristics, and the narrow-band nature of analog circuits and the low bandwidth of prior art digital systems ensured that the low-pass characteristics was present in spite of the fact that the equivalent series resistance (ESR) and the equivalent series inductance (ESL) values of the bypass capacitors were relatively high.
In modern electronic circuits, the target impedance is so low (typically in the milliohm range to up several hundred hundred megahertz frequencies), that it is not practical to rely on a power-distribution structure consisting of individual inductive components (inductors, wires, traces) and individual bypass capacitors to suppress noise. One prior art approach to produce a low target impedance is to utilize entire, unbroken planes in the multi-layer boards to deliver power and to provide ground. In addition, in order to increase the bypass capacitance between the signal traces and the power and ground planes, one approach in the prior art is to use a material with a high dielectric constant (Dielectric constant) in the insulating layer between the signal traces and the power and ground planes. The high dielectric constant material allows charge to be stored locally in the insulating material in order to provide a bypass capacitance function.
However, in order for electronic circuits to operate at high frequencies, the traces which carry digital signals from one component to another, called signal interconnects, must have a short propagation delay and have low signal losses at high frequencies. The signal propagation delay in a signal interconnect is primarily influenced by two factors: the interconnect length and the dielectric constant of the insulating material in the insulating layer between the signal interconnect and the power and ground plane. To make the propagation delay shorter, materials with a lower Dielectric constant must be used. Losses in the signal traces come from both the copper loss and dielectric loss and, to minimize the high-frequency losses, the dielectric constant of the insulating materials must also be low. However, the use of low dielectric constant materials reduces the locally available charge and, thus, conflicts with the noise reduction requirements of the power distribution network.
In addition, the use of insulating materials with the low dielectric loss together with the large planar power and ground layers create electrical resonators, and both the self and transfer impedances of such a structure exhibit an infinite number of resonance peaks at different locations in the structure. Consequently, the power supply impedance is increased at these peaks and noise can be transferred from one point to another with less attenuation and even amplified far from the noise source. For example, FIG. 1 is a perspective view of a pair of square conductive planes 100 and 102 separated by a dielectric layer 104 comprised of a fiberglass-epoxy composite material called FR4 which is in common use for constructing printed circuit boards. Each conductive plane has a side length (L) of 10 inches and is made of copper and is 0.0014 in. thick. The FR4 insulating layer separating the planes has a dielectric constant of about 4.7 and has a thickness (h) of 0.002 inches. FIG. 3 is a graph of the magnitude of the simulated electrical impedance between the pair of rectangular conductive planes of FIG. 1 (log.sub.10 scales) versus the frequency of a voltage between the planes (log.sub.10 scale). The graph illustrates the "self impedance" taken at three different locations: at the center of the structure (MSCntr), at one corner (MSCrn) and at the middle of one side (MSMidx.) The graph was created by modeling each of the pair of conductive planes as a grid of transmission line segments 200 connected at nodes 202 as illustrated in FIG. 2. The grid size was one inch and the impedance value was simulated by stepping a unity magnitude swept-frequency current source through all of the nodes 202 and calculating the voltages at all of the circuit nodes. The "self" impedance at any node is the complex voltage at the node while the current source is also at the same node. The transfer impedance between a first node and a second node is the complex voltage at the second node while the source is located at the first node.
As shown in FIG. 3, the magnitude of the self impedance between the parallel conductive planes 100 and 102 of FIG. 1 varies widely at frequencies above about 100 MHz. The parallel conductive planes exhibit multiple electrical resonances at frequencies between 100 MHz and 1 GHz, resulting in alternating high and low impedance values that affect the power supply impedance and noise transmission characteristics at these frequencies.
It would thus be desirable to have a system for power distribution which had no resonances within the bandwidth of the signals, had high attenuation for noise propagating between locations and had high charge reservoir capability to reduce power supply impedance without increasing the signal propagation delay.